Claude code, reasoning models, useful agents, 2025 was a wild year...

Thank you Blacksmith for sponsoring! Check them out at: https://soydev.link/blacksmith

SOURCES
https://simonwillison.net/2025/Dec/31/the-year-in-llms/
https://www.greptile.com/state-of-ai-coding-2025
https://www.youtube.com/watch?v=AmdLVWMdjOk
https://steipete.me/posts/2025/shipping-at-inference-speed
https://x.com/steipete
https://x.com/simonw

Want to sponsor a video? Learn more here: https://soydev.link/sponsor-me

Check out my Twitch, Twitter, Discord more at https://t3.gg

S/O Ph4se0n3 for the awesome edit 🙏

T3Dotgg's resource offers insights, tutorials, and resources for gamers and esports enthusiasts. Readers can learn about gaming news, esports events, and professional gaming strategies. With articles, reviews, and interviews, T3Dotgg provides  guidance and expertise for staying informed and competitive in the world of gaming.

Theo - t3․gg

AI coding tools transformed software development in 2025, with reasoning models, coding agents like Claude Code and Codex, and terminal-based workflows becoming mainstream. Developers now write significantly more code overall while spending less time manually editing, with PR sizes increasing 33% and lines per developer up 76%. Models can now complete tasks taking humans 5+ hours, Chinese AI labs gained prominence with competitive open-weight models, and "vibe coding" evolved from novelty to practical workflow. The shift includes $200/month subscription tiers becoming standard, YOLO mode enabling autonomous agent execution, and developers increasingly coding from phones via asynchronous agents.

2025: The year I stopped writing code

I noticed that while it’s more effective time wise to use copilot to write code, but there are a couple of caveats:
<ol>
<li>After a couple of weeks of only generating code I found out that my problem solving skill becoming worse, im becoming lasier, and instead of thinking I prefered to delegate to AI. In a long run obviously I will become dumber</li>
<li>If I’m in hurry and not checking implementation of AI, it would lead for me not to understand part of a codebase. Obvious solution is to read deeply and understand all the logic, but it takes about the same amount of time if I would have written everything myself. And in a long run it would mean that control over codebase is lost, meaning more time to debug and fix if AI wouldnt be able to do that.</li>
<li>When I’m overrelying on AI I stopped enjoying work, it’s not actually my work anymore, i’m just a simple PC operator that gives commands.</li>
</ol>
When I realized that i’m losing myself because of AI, I decided to start writing solution to most problem by myself, only delegating simple tasks to AI. Not I’m enjoying my work again

Quite wild. Sums up the year pretty solidly. <s>Although for me I still wrote lots of code, the difference is that now I am editing code instead of writing the draft</s>